Idea Map of Sprited

The project ideas so far.

• Pixel AI Agent (on-going): Company’s internal AI agent. Later to be a public facing.

• Machi(on-going): Multi-agent simulation ground where AI agents are given (virtually) physical form and can live in on a shared world. “Machi“ means “town“ in Japanese. It also embeds first five letters of “Machine.“

• Human-Like Memory System: Memory system the mimics human brain memory system. The premise is that memory is that forms the unique self identity.

• Side-kick Model for Grandma Models: Current pre-trained LLM models are like very wise almost omniscient grandparent. However, their short-term memory is impaired. We build a short-term memory module that is equivalent to “RAM“ in a PC. Don’t use just chat history or RAG, but actually build out a efficient model that can efficiently store short term memory. It can later be extended to long-term memory.

• Memory as a Service: Imagine you can type into ChatGPT and have the same memory available in Claude chat.

• Tinker AI Agent (on-going): Ideation of company’s AI agent that is fully autonomous inside container environment. The idea is that it will be replicating what the main developer is doing and accelerate it own its own. The main job of this agent will be create Pixel.

• Hidden Market that’s between Large LLM walled garden offerings: ChatGPT, Anthropic and other companies are competing. The will all try to setup their own “walled garden.” This provides unique market where if you are able to be a “brokerage“ between them. Or information sharing agent between them will be a very lucrative market.

  • A “bridge” is needed between these agents. Some of the agents are way to focused, and needs help from other agents to course correct.

• Data Collection & World Knowledge Archive - Pivoting the company to Data Crawler: Center the company’s focus around crawling, gathering generating and automatizing data gathering process. This will basically collect infinite amount of data. And cover all the ground to be used for various model trainings.

• AI of AIs: AI model that sits on top of other AI providers and automate things at a meta level. Think of hyper parameter tuning over various different aspect of orchastration.

• Meta-Cognition AI: AI node that serves as meta-cognition loop. Imagine Github Copilot with ability to reason about “can I really do this task?“